MSG-GAN Based Synthesis of Brain MRI with Meningioma for Data Augmentation
S. Deepak, P. M. Ameer
Abstract
Deep learning techniques have found significant applications in the development of computer aided diagnosis (CAD) system for brain tumor characterization. The performance of such a system is often affected by the limited availability of brain MRI images with a specific tumor, in the form of training data. To an extent, this limitation can be minimized by the use of synthetic images produced using generative adversarial networks (GAN). In this paper, a multi-scale gradient GAN (MSG-GAN) is used to synthesize MRI images with meningioma disease. The synthesized images are used to augment the training set of a multi-class brain tumor classification problem using a deep convolutional neural network (CNN). The evaluation is done on the coronal-view images from the figshare database and it resulted in an improvement in the classifier's performance, in terms of the balance accuracy score.